ALT A breakdown of groups susceptible to harm at every stage of the AI dataset and model development stage. Table from a forthcoming research paper (Title: The Subjects and Stages of AI Dataset Development: A Framework for Dataset Accountability. Authors: Mehtab Khan, Yale Law School, Harvard University & Alex Hanna, Distributed AI Research Institute). First column, Stages of development (problem formation [PF], data collection [CO], data cleaning [CL], data annotation [DA], model training and evaluation [TE], model implementation and inference [MI], data and representation distribution [DR]). First row, Who is implicated (data subjects [DS], data annotators [DA], copyright holders [CH] and model subjects [MS]). PF corresponds to DS, DA, CH, MS. CO corresponds to DS, CH, MS. CL corresponds to DS, MS. DA corresponds to DS, DA. TE corresponds to MS. MI corresponds to DS, CH, MS. DR corresponds to DS, CH, MS.